ETHICAL, CLINICAL, AND REGULATORY CHALLENGES OF USING LARGE LANGUAGE MODELS FOR CLINICAL DECISION SUPPORT IN MEDICINE: A COMPREHENSIVE ANALYSIS

• Dilbar Komilova

  Student of Tashkent State Medical University,Tashkent Uzbekistan

  Author

• Fazliddin Arzikulov

  Assistant, Department of Biomedical Engineering, Informatics, and Biophysics, Tashkent State Medical University, Tashkent Uzbekistan

  Author

Large language models (LLMs) are increasingly incorporated into clinical decision‑support (CDS) systems, offering rapid synthesis of medical knowledge, automated documentation, and data‑driven insights. However, their deployment raises intertwined ethical, clinical, and regulatory challenges. Ethically, LLMs can propagate bias, generate convincing yet inaccurate content, and threaten patient privacy, prompting calls for transparent oversight and human‑in‑the‑loop governance[1]. Clinically, performance evaluations reveal that state‑of‑the‑art LLMs often underperform physicians, fail to follow diagnostic guidelines, and are sensitive to prompt framing, limiting reliable integration into care pathways [2][3][4]. From a regulatory perspective, existing medical‑device frameworks struggle to classify LLM‑driven CDS, creating uncertainty around FDA compliance, post‑market surveillance, and liability [5] [6]. This thesis conducts a systematic review of peer‑reviewed literature (2018‑2024) and semi‑structured interviews with clinicians, ethicists, and regulators to map these challenges and propose a multidisciplinary framework. The framework emphasizes (i) fairness and bias mitigation, (ii) rigorous clinical validation and explainability, and (iii) adaptive regulatory pathways that treat LLM‑based CDS as hybrid software‑medical devices. Implementing such safeguards can reconcile innovation with patient safety, equity, and legal accountability.

• pdf

This work is licensed under a Creative Commons Attribution 4.0 International License.